Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
NBER WOR~G PAPER SERIES
STEPPING STONE MOBIL~Y
Boyan JovanovicYaw Nyarko
Working Paper 5651
NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue
Cambridge, MA 02138July 1996
This paper was prepared for the Carnegie-Rochester Conference Series on Public Policy. We thankthe C.V. Starr Center for Applied Economics at New York University and the National ScienceFoundation for financial help. We also thank Lanny Arvan, Will Baumol, Steve Davis, DouglasDwyer, Ron Findlay, Jeremy Greenwood, James Heckman, Barbra Heyns, Jacob Mincer, OliviaMitchell, Silvio Rendon, Stephanie Wilk, Nachum Sicherman, Aloysius Siow, and WilbertVanDerKlaaw for comments, This paper is part of NBER’s research program in Productivity. Anyopinions expressed are those of the authors and not those of the National Bureau of EconomicResearch.
@ 1996 by Boyan Jovanovic and Yaw Nyarko. All rights reserved. Short sections of text, not toexceed two paragraphs, may be quoted without explicit permission provided that full credit,including @ notice, is given to the source.
NBER Working Paper 5651July 1996
STEPPING STONE MOBILITY
ABSTRACT
People at the top of an occupational ladder earn more partly because they have spent time on
lower rungs, where they have learned something. But what precisely do they learn? There are two
contrasting views: First, the “Bandit” model assumes that people are different, that experience
reveals their characteristics, and that consequently an occupational switch can result. Saond, in our
“Stepping Stone” model, experience raises a worker’s productivity on a & task and the acquired
skill can in part be transferred to other occupations, and this prompts movement. Safe activities
(where mistakes destroy less output) are a natural training ground.
Boyan JovanovicDepartment of EconomicsNew York University269 Mercer Street, 7th FloorNew York, NY 10003and NBER
Yaw NyarkoDepartment of EconomicsNew York University269 Mercer Street, 7th FloorNew York, NY 10003
1. Introduction
“You must be able to walk before you can run.”
The claim is that one must learn the simple before mastering the difficult. It clearly is true in the
educational process, where learning a subject like mathematics starts the student off doing simple
exercises, and then moves him onto more and more complex problems. It is equally true in the
training process -- whether a person is learning how to play a musical instrument, how to drive a
car, or how to operate a machine. But the claim also seems to hold as a description of the
sequence of activities a person does over his or her entire working life: simpler tasks and j obs
tend to precede more complex and difficult ones. Using several different measures of job
complexity, Wilk and Sackett (1995) find that as people age, they tend to perform harder tasks.
And so it would seem that the principle that one begins with the simple and then moves to the
complex, should also apply to how people choose the sequence of jobs and occupations over
their lifetimes. That is the claim of this paper: That occupations are like rungs on a ladder --
that they are, in other words, like “stepping stones”.
The opening phrase also takes as given the obvious: walking is a very different activity
from running. There are many examples of distinct activities that makeup a career path, and
Table 1 lists some (most of which we owe to Nachum Sicherman), In many, one ~ be in one
in order to get to the next, while others describe typical careers. Some are likely to take place
within one firm, while others can occur only across firms:
---------------------------------------------------------------------------------------------
Professor + Dean Actor + President Police officer + Security expert
Player + Coach Engineer + Manager Lawyer + Politician (senator)
Lawyer + Judge Police officer + Private Detective Team member +Team leader
TA/RA + PhD White Belt + Black belt Priest + Cardinal
Soldier + General Dancer + choreographer IRS employee +
Associate + Partner Kitchen worker+ Chef Tax consultant
-----------------------------------------------------------------------------------------------
Table 1: Examples of Occupational Ladders
Some would argue that occupations exist merely to reward better workers with
“promotions”, and that to focus on occupational ladders is to look at the surface, the veneer, and
that this can only serve to hide the deeper forces underneath. We disagree on two grounds: First,
an occupational change usually signifies some shift in the kind of work being done. For
example, on a company’s job ladder, the person at the top does ti do the same thing as a person
at the bottom. Occupational distinctions are thus more than labels. And second, the
occupational ladder surely & a symptom of a deeper force -- human capital accumulation -- and
studying how people move up the ladder cannot but improve our understanding of this more
fundamental concept.
Climbing the occupational ladder ofien involves changing firms. Twenty years ago, Hall
and Kasten noticed that most job-changers move to a higher-paying 3-digit occupational
category. Since average pay in an occupation is widely known, why don’t people enter the top
occupations right away instead of starting with the lower paying ones? And does the answer
have something to do with the tendency for people to undertake more complicated tasks as they
gain experience?
The answer, of course, is that people at the top of the ladder earn more because they have
spent time on lower rungs. Our model says that jobs that are “close” in information content will
form rungs a ladder, but the safer ones should come m : On such jobs, mistakes destroy less
output, and hence they are a natural training ground. A person must “work up” to being CEO --
or else we would expect them to make costly errors. Secondly, if mistakes are costlier on
complex jobs, labor should flow from occupations with flatter learning curves and into
occupations where learning curves are steep.
A different hypothesis for why a worker moves from job to job is that the accumulation
of experience is accompanied by a sorting process in which employers and workers learn what
skilled job each worker can do best. As they learn it, the worker is assigned to occupations
where his kind of ability is needed. This is a model not of an accumulation of knowhow, but of
the accumulation of information about a worker’s innate traits -- his or her comparative and
absolute advantage. We refer to this hypothesis at the “Bandit model”.
We shall argue, however, that the Stepping Stone model is the natural job-ladders model,
because it emphasizes growth of skill on one job d its subsequent transfer to another, better
job. This notion is central to the models of Weiss (1971), Rosen (1972), and Sicherrnan and
Galor (1990). Far from being an unpleasant surprise, they argue, occupational mobility can be a
- career move -- the worker earns more lifetime income by taking up more than one
2
occupation in sequence. We extend the work of these authors by providing a microfoundation
for the notion of general skill, skill that is usefil in a variety of endeavors, and we then derive
some further implications of this view. Related information-theoretic models are Prescott and
Visscher (1980) and O’Flaherty and Siow (1992). But these models are about the firm screening
the worker, and not about the transfer of acauired skill.
Human ca~ital. The term “human capital” has now come to mean so many things that it
is too general for our purposes. Even the distinction between innate and acquired skill is too
vague unless given specific content. Any acquired job-specific skill is human capital, but then so
too is information about oneself or about features of a job or industry that may affect one’s
prospects there. Any piece of information that can be used to raise a person’s lifetime earnings
in a sense augments a person’s stock of human capital. In a world in which people and jobs
differ and in which these differences are not immediately and commonly known, one must be
more specific about what human capital is, at least when asking questions about occupational
choice and the behavior of earnings and other aspects of labor market histories.
We argue that different kinds of information acquisition have very different impacts on a
person’s labor market history, and we distinguish information into two types:
(a) information about personal traits and about the content of jobs, and
(b) information about how to perform a given task.
The first kind of information (a) allows a person to choose a job where he fits best. Sometimes it
is the employer who will assign the worker to some position and who therefore needs to know a
worker’s ability which, together with the job requirements will determine whether the worker is
suitable for the position. This is the kind of information that drives the Bandit model. The
second kind of information (b) is the kind that augments the skill of a given person on a given
task -- learning how to type, how to drive, how to cook, how to play tennis. This is the kind of
information that drives the Stepping Stone model.
Our aim is twofold. First, we want to understand occupational mobility. But second, we
want to use this exercise to deepen our understanding of the general concept of human capital by
distinguishing (a) from (b), which we see as findarnentally different. And we shall show that not
only do (a) and (b) differ conceptually, they can also have sharply conflicting implications.
Section 2 presents our version of the Stepping Stone model. Since it is important to
clarify the differences between the different motives for labor movement, we briefly describe the
sectoral shock model in section 3, and then the Bandit model in more detail in section 4. Section
3
5 lists some facts that help us discriminate between the models. Section 6 briefly indicates how
the Bandit model and the Stepping Stone model can be combined into a hybrid. Section 7
provides some further discussions and observations, section 8 concludes.
2. The Stepping Stone Model
We begin with a partial equilibrium context. The appendix shows how it fits into a
general equilibrium setting.
2.1. Production possibilities and the information structure.
A farmer lives for two periods. In each period, he can grow either apples or bananas, but
not both. The production function for apples is
q=l-(x-zA)2,
where qA is the number of apples grown, z* is a decision the farmer must take, such as the
amount of irrigation and fertilizer to apply, and x is a random variable, such as the weather,
realized only after z* has been chosen. The production function for bananas is
q~ = [1 - (y - ZB)2],
(1)
(2)
where q~ is the number of bananas grown, z~ is again a decision, and y is a random variable.
A multiplicative normalization constant can be added to this production function, but it would
leave all of our results unchanged. The random variables (x, y) have a bivariate normal
distribution. The unconditional variances of x and y are OX2<landoY2<l. Letp=
uX,Y2/0X20Y2 denote their squared correlation coefficient. Then the conditional variances are
Var{y I x} = (1 - p)(s;, and Var{x I y} =(1 - p)u~, (3)
and they do not depend on the values of x and y that are realized.
One period of work growing apples reveals x, and one period growing bananas reveals
y. It also reduces the variance of the other random variable to its conditional variance. A person
can not learn x and y by talking to or watching other farmers -- hands-on experience is both
necessary and sufficient for learning to occur.
4
Let p be the price of bananas in terms of apples. We shall only consider equilibria in
which p is a constant. Formally, this is valid if the x’s and y’s of agents are all independent of
each other. In this case the strong law of large number implies that the per-capita output of
apples and bananas will be constant over time, and that prices will actually be constant. The
Appendix lays out an overlapping generations structure formally.
We shall measure the rate of interest, r, in units of the price of apples at different dates.
Since p is constant, this rate will be the same as the banana-denominated interest rate, and will
satisfy (1 + r)- 1 = 5 (the discount factor for second-period utility). We assume that the farmer
will maximize ex~ected discounted lifetime income. (In the appendix we assume preferences
that imply an indirect utility fiction linear in lifetime wealth. These preferences assume that
consumption occurs only in old age. Also, productivity rises as people age. So for prices to be
constant over time, we need a stable demography, and so we assume that there are overlapping
generations).
In each period, a farmer faces an occupation choice (A or B), and a production choice (the
level of z), which must be set before x and y are observed. The risk-neutral apple producer
will set z* = E(x), and the banana producer will set z~ = E(y). These expectations are
conditioned on the information that the producers have at the time. Since their first-period
decisions are set on the basis of no signal at all, their expected outputs (measured in units of
apples) from following this strategy are 1- 0X2,and p(l - 0Y2),respectively.
There are four possible career paths for the farmer: AA, AB, BB and BA. The expected
discounted lifetime incomes from these four paths are
v AA = 1-(JX2 +6, ‘AB(P) = 1 - ~: + bp[l -(1 - p)cryz],
v BB = P[(l -~;) + ~1> and V~A(p) = p(l - UY2)+ 5[1 - (1 - p)ox2],
The Vi,j are perfectly foreseen, because Var{y x} and Var{x I y} are not random.
Everyone is the same ex ante. There is no innate absolute advantage, no innate
(4)
comparative advantage. [The latter are the essence of the Bandit model, but they are entirely
absent here.] The Stepping Stone model is one of learning how to ~ something, and not a
model of learning about oneself or learning about the quality of the worker-job match.
5
Our argument revolves around p. This is an index of how general human capital is.
While VAA and V~~ do not depend on it, VA~ and V~A are strictly increasing in p. This
result reflects the role that p plays -- it is an index of how much human capital can be
transferred between apple growing and banana growing. Indeed, when p = 1, p = 1 and OX*=aY2,
apples and bananas become the same good as far as production is concerned.
Proposition 1: Suppose that aY2 > OX*. Then V~Ais always dominated, regardless of the
value of o and ~,
Proofi when p> 1, V~~>V~A and when p< 1, V~A>V~A.~
In our 2-occupation economy, p is an index of how general human capital is, because it
is a measure how much one can learn about x from observing y and vice versa. In an economy
with many occupations, there will in general be a different p for each pair of occupations, and
afier choosing his first j ob, the worker would in general face a different p for each possible
occupational change. But as long as there are just two periods of life, Proposition 1 would
continue to apply.
So an agent will never move from a high variance occupation to a low variance one. The
intuition for switching emerges from the following special case: Suppose p = 1. Then learning
x means that we also learn y. But suppose OX2 is much smaller than 0Y2. Then the
experimentation cost (in terms of foregone output) of learning y is much less if the young
person produces apples. We shall now derive a general result formalizing this intuition. Define
the following fmctions of p:
Po(p) =
PI(P) =
1-OX*
and
P2(P) ❑
1
1 - (1-p)u:
6
Then one can easily verifi the following:
(i)
(ii)
(iii)
v AB ~ VAA as p ~ PJ(p) ;< <
> >v BB =VAA as p=pl; and
< <
v BB ~ VAB as p ; Po(p) .<
We provide a diagram of the functions Pi for i = 1,2,3, which uses some properties that
we collect together in the lemma below:
Lemma 1: Suppose 0Y2>oX2.Then (i) PO(p) (resp. P2(p)) is strictly increasing (resp. strictly
decreasing) and continuous in p; (ii) PO(0)< p, < P2(0); (iii) P2(1)=1 < P, < PO(l); and (iv)
there exists a pi in (O,1) such that
Po(pc) = P,= P2(PC)
where
6a:+o~(l-lJ:)p, =
0Y2(1-0X2+ 5)
Proof Consists of easy algebra. ■
Figure 1 uses the facts in the lemma to depict the functions Pi, indicating the optimal
career paths in p - p space. Figure 1 shows the following: When p < pC, there is no
occupational switching, regardless of p. When the price is higher (resp. lower) than PI agents
spend all of their lives growing bananas (resp. apples). When p is equal to PI , agents are
indifferent between spending their entire lives growing apples or growing bananas.
Now suppose that p > pC. Then when p is very high (resp. low) it is optimal for agents
to spend all their lives in banana (resp. apple) farms. But when p is at an intermediate level, in
particular when p e [P2(p), PO(p)] it is optimal to switch professions: agents will grow apples
when young and bananas when old. So with the aid of Figure 1, we have proved
7
P,(o)
P,
P,(o)
o
\ \ \ Po (P)\\\\\ B+B\\.‘.‘.,......--------..-”,.-,-P2(P)
ATA
Pc
Fi~ure 1: Optimal Career Paths in p-p Space
Proposition 2: Occupational movement can occur only if p > p, .
So, occupations will form adjacent rungs on an occupational ladder only if human capital
is sufficiently transferable between them. Note too that while the condition p > pC is only
necessary, it is also sufficient if in addition both goods are supplied in equilibrium. This is clear
from figure 1 --if p > pC and we are outside the A + B region, we must be in a region where
only one good is produced.
Let us use the notation i+j to refer to an agent who chooses farm i when young and farm
j when old where i, j = apple, banana. B+B agents produce respectively 1-OY2and 1 units of
bananas when young and old. The A+B agents produce respectively Oand 1-(1 -p)a~ units of
bananas when old. From Proposition 1 we there will be no B+A. By assumption there is a unit
mass of generation of agents. Fig. 2 depicts the supply curve of bananas for the regime p < pC.
8
This supply curve is obtained by selecting a particular value of p that is & than pC,
and then varying p while holding p constant, In the figure, we use the definition V* E Max
{VA*, VBBY ‘A., V~A} . This figure shows that supply responds much as it would in a static
model; above some threshold price P, , everyone would wish to grow bananas, The equilibrium
number of farmers of each type will then depend on the demand for apples and bananas. The
general equilibrium aspects are detailed in the appendix.
Similarly, Figure 3 depicts the supply curve for the regime p > pC. This supply curve is
obtained by selecting a particular value of p that is greater than pC, and then varying p while
holding p constant. Behind this supply curve is a shifiing nature of switching behavior, as a
function of p. Figure 4A depicts the fraction of the young workers that are switching. When Q~
= O, all the young are in apples, and no one ever switches. Similarly, when Q~ = 2- UY2(the
maximal output of bananas), there is again no switching, since the young and the old are growing
bananas. Movement is the largest at Q~ = 1 -(1 - p)o$. At that point all the young are
switching from A to B! To the lefi of that point, all the young are in A, and only some switch
to B, and to the right of this point, only some of the young are in A, and they all later switch to
B.
Figure 4B shows there is a whole range of consumer utility functions leading to a whole
range of demand functions and resulting equilibrium prices p for which all the young start off in
A and then they M move to B, This is not an occurrence of zero measure in parameter space.
9
P
P,
o
V*=VAA=VBB
V“=VAA
V“=VBB
2 -0,2
Fig. 2: the Supply Curve of Bananas men ~s o:.
P
<PO(P)
)2(P)
0
V*=VAA=VAB
V*=VAAI
1
V“=VAB=VBB
V*=VAB
V“=VBB
l-(1 -p)u; 2-0/
Fig. 3: The Supp lV Curve of Bananas When 0> DC
11
Fractionof youngthat movefrom A+B
o 1-(1 -p)ayz 2-0Y2 ‘B
Fig. 4A: Step~in~ Stone Movement: Fraction of Young That Move as Function of Out~ut
1 ‘-’- ‘---” ----
Fractionof youngthat movefrom A+B
o l’2(P) PO(P)P
Fig. 4B: Ste~~in~ Stone Movement: Fraction of Young That Move as Function of Price
12
2.2. Age-emin~s profiles andpromotions inthe Stepping Stone model.
The Stepping Stone model (and the Bandit model of section 4) both assume that career
choices are made by the worker. The models describe workers’ supply -- which jobs they are
-g to undertake, in what sequence. Every job is open to every worker, and it is the wage
structure that induces the worker to make the equilibrium choice of j ob. In reality it often is the
h that decides what task a worker will perform. The firm often actively trains a worker,
evaluates the worker’s abilities, and then assigns him to whatever task it deems is appropriate for
that worker. How consistent is the model with this observation?
There is an interpretation in which firms enter at zero expected profit, organize tasks A
and B under one roof, allocate workers to tasks A and B, and pay wages. Under this
interpretation, ~ is the number of efficiency units of service A that the worker supplies, and
q~ is the number of efficiency units of service B. Then p is interpreted as the price per
efficiency unit of service B, and the relative price per efficiency unit of skill A is normalized at
one. Wages could either be piece rates (so that the worker gets q* or q~) or time rates, in which
case the worker would get his expected output conditional on his experience. Profit-maximizing
behavior would dictate that the firms’ allocation choices coincide with the equilibrium choices
described here.
The deterministic approach to human capital has emphasized the growth of wages with
job-tenure and with labor market experience. Typically, one specifies expected output as a
function of experience, and then adds a disturbance with no specific rationale. Our theory gives
rise to implications about all the moments of wages, and hence it complements the traditional
approach. To bring this point out more clearly, Figures 5A, 5B, and 5C describe lifetime
expected earnings in the three types of equilibria in the Stepping Stone model that can occur
when p > p, . Which one prevails depends on where exactly p is.
13
Average earnings in B
P(l-(1-P)UY2)
1-0: Im
Average earnings in A
AGE1 2
Fia: 5A. D E (P7(0). PO(0)] -- All Start in A. All Move to b
$
1Average earnings in B
.1 L
1-0,2
Average earnings in AI
1 2 AGE
Fig: 5B. D = P2(p) -- All Start in A. Some Move to b
$
1Earnings in
P---- ---- -
P(l-(1-P)~y2) Earnings in
l-a:
B of B-young
B of A-young
t
P(l-a:) ------=Earnings of A-young
Earnings of B-young1—
1 2 AGE
Fig: 5C. u = PO(O) -- Some start in A. some start in B, all end u~ in B.
14
3. The Sectoral Shock Model
Suppose the automobile sector makes cars using engineers and manual labor as inputs. If
an increase in the demand for cars causes an inflow of labor of both types into the automobile
sector, we will see a net inflow of labor, and the gross flow of workers will equal the net flow.
In fact, however, gross flows among sectors, both of jobs (Davis and Haltiwanger 1992, Table
1) and of workers (Jovanovic and Moffitt 1990, Table 1) are much bigger than the net flows.
To explain this with a sectoral shock story requires that the shock -- say the invention of a new
mass-production technique -- increase the demand for engineers, and displace the unskilled
workers. (Krusell et al (1995) point out that the falling relative price of equipment may have
caused the skill premium in wages to rise in recent years). We should then see an inflow of
engineers, an outflow of unskilled workers, and therefore the gross flow will exceed the net flow
of labor between the automobile sector and the rest of the economy.
This refinement of the sectoral shock still does not explain occupational ladders. In an
occupational choice equation, sectoral shocks explain “time dummy” effects, not “age” effects.
The latter are the focus of the Stepping Stone model.
4. The Bandit Model of Occupational Choice
What follows is a two-period version of the model analyzed first by Johnson (1978), and
its central implication, as described by Miller (1984, p. 1088) is “that the inexperienced
gravitate towards risky ventures. Because younger workers are less experienced than their older
counterparts, they are more willing to try jobs where success is rare; older people who have quit
them already know they themselves are unsuitable.”
Recall that in the Stepping Stone model, such behavior is impossible. In the Bandit
model, it happens for some, if not most parameter values. However, as suggested by an example
in Waldman (1984), we shall see that it does not happen for ~ parameter values.
4.1. The 2-period Bandit model.
Let there be two occupations, A, and B. Let a and ~ denote a person’s productivity
in occupations A and B respectively. As in Roy’s (1951) model, there is a distribution of the
vector (a, ~) in the population. Both Johnson and Miller assume that this distribution is
normal:
15
Let
Cov2(a,~)r=
0.2UP2
bethesquared correlation coefficient, Weshall repeatedly usethe following result:
Lemma 2: Under the above assumptions,
E(al~) = Ea + c(~-E~) where c = [Cov (a,~)/o~ ]
E(~la) = E~ + d(u-Ea) where d = [C”v (u,~)/0a2 ]
Var (a I~) = (l-r) 0=2 and Var (~la) = (l-r) UP2
Proof [Chow (1983, p, 12)] ■
In the appendix we also show that when a. 2< 0P2,the parameters can generate any c in (-
1,1) (and nothing else) and any d ~ R.
At the outset, agents know the population distribution of (a, ~), but not their individual
(a, ~) type. A lifetime is just two periods here, so we assume that afier a period an individual
learns com~letelv his productivity in the occupation he experienced. If a and ~ are correlated,
this will allow him to infer something about his productivity in the other occupation. For
example, a young person in occupation A would, after one period, learn his a. He then will
revise his prior distribution over ~ to be the normal distribution with mean E(P Ia) and
variance Var (~ Ia), as given in Lemma 2. Note that just like the Roy model, this is a model
where some workers have a comparative advantage in one occupation, and some in the other.
But some workers also have an absolute advantage over others. Nevertheless, this is a model in
which assignments are based entirely on comparative advantage, and we shall have more to say
about this in section 7.1.
16
Assume that the individual is risk neutral, and that he discounts second period income by
the factor 6. Suppose that his earnings equal his productivity. Further, let the price of the skill
~ relative to that of a be unity. (If it were not, we could define a new ~ whose mean and
standard deviation equal that of the old ~, multiplied by the price of that skill, and the analysis
applies). Then a person with u > ~ has a comparative advantage as an apple grower, and a
person with a < ~ has a comparative advantage as a banana grower. The costs of switching
occupations are assumed to be zero.
4.2. The lifetime decision ~roblem.
If, when young, the individual worked in occupation A, then he will have learned a by
the end of the period, and he will switch occupations if and only if u < E[~ Ia]. Otherwise he
will remain in occupation A, where he will earn u in the second period. Let V= be the
expected present value of starting out in occupation A. Then,
V= = Ea + 5E[Max {a, E[~lu]}] .
By a symmetric argument, the expected present value of starting out in occupation B is
Vp = E~ + 5E[ Max {~, E[ul~]} ] .
The following result characterizes the first period optimal decision:
Proposition 3: For each value of r there exists a critical number A’(r) such that
Vp : vu>
as E~ - Ea ~ A’(r).
Further, A’(r) is strictly negative for r in [0,1); AC(l) = O; and AC(r)is continuous in r.
Proofi See appendix. ■
(5)
(6)
This is illustrated in fig. 6 below.
17
o
Region where B is chosen first
I Typical Johnson (1978) parameter vector
r
w
\Typical Waldman (1984)
parameter vector
Fig: 6. The Optimal Date 1 decisions in ((EB-Ea) . r) space
18
Wecannow prove Miller’s assertion. Assume thatthe meanpayoffin thetwo
occupations is the same, but that occupation B is riskier. Then the young worker will indeed
prefer to start out in occupation B. [This of course is a direct implication of Proposition 3 (by
setting A~E~ - Ea = O).]
Corollary 1: (Johnson 1978, Proposition 1 -- The young prefer the riskier occupation). Suppose
that Ea = E~ , 0=2< aP2and r<l. Then V.< VP .
We now have, for the second period decision,
Lemma 3: Unless Cov(a, ~) = a: or Cov(a, ~) = 0P2 , the inequalities
both occur with positive probability. That is, the inequalities are both met generically in the
parameter space.
Proof The marginal distributions of a and ~ are normal, and so all points on the line have
positive density. Then by Lemma 1, it is clear that there will exist some intervals on the line
where each of the inequalities mentioned in the statement of this lemma will hold. Since a
positive probability will be assigned to these intervals, this lemma is tree. W
Lemma 3 implies that conditional on there being some young workers in both
occupations, after a period of experience some of the labor will flow in both directions. But if
young people are all the same ex ante, they must be indifferent between the two occupations.
That is, we must have V= = VP. When this is true, the model gives rise to flOWS of labor in
both directions:
Proposition 4: (Occupational switching in both directions). Suppose that VW = VP. Then
generically in the space of parameters {UU2, OP2, COv(a, ~) }, labor will flow in both
directions between the two occupations.
Proof The additional restriction V= = VP does not imply either equality in the first line of
Lemma 3 ■
19
From Lemma 2, we know that if we consider E[a I~] as a finction of ~ then a graph of it
against ~ will have slope c with Ic I<1. Suppose that Eu=E~=l. Then the point of intersection of
such a graph with the 45 degree line is at ~=E[u I~]=1. We therefore obtain figures 7A and 7B:
4.3. A~e-earnin~s profiles and promotions in the Bandit model.
A potential difficulty with the Bandit model as a model of occupational ladders stems
from Corollary 1, and from the fact that top-paying occupations also have a higher variance of
earnings. Suppose B is the high-variance occupation, so that 0P2 > oa2 . If “top-paying” is
interpreted as a statement about unconditional means so that E(p) z E(a), then Corollary 1
implies that all young workers would want to start at the top, and this is counterfactual. Indeed,
suppose that Ea = E~ = 1 so that all the workers begin in B. It should be clear from fig. 7A
that those who move from B to A do worse than those that remain in B in the second period.
In other words, those that move would appear to have been “demoted,” when we look only at
second period wages. This is illustrated in fig. 7B. A further problem with this explanation of
occupational mobility stems from Corollary 1: B is the riskier activity, and so the variance of
earnings over people shrinks dramatically over the life cycle. But in fact this variance grows
with labor market experience (e. g. Mincer 1974, 0 ‘Neill 1995), It is easy to show that in the
Stepping Stone model, if p is sufficiently less than 1, variance of earnings in the cohort of
workers can rise over the life cycle, directly as a consequence of Corollary 1-- safe activities are
chosen first.
However, “top-paying” is a statement about the earnings of those workers that end up
choosing B. In this case it is possible that E(p) < E(a), and Corollary 1 no longer applies.
Indeed, as suggested by the Bandit model analyzedbyWaldman(1984), it is possible to find
parameter values in the Bandit model under which
(i) the safer occupation A is sampled ti, that
(ii) people that switch from A to B raise their earnings in the process, and
(iii) those who switch from A to B earn more than those who remain in A.
Such predicted behavior is what one would associate with a career that begins in A, and ends in
B for those who are “promoted”. Inspecting fig. 6 shows that the following easily verified result
(which implies (i) - (iii) above):
20
Agents who Switch to .~Agents who Stay in B
-
—
9- -Aver(n-- -- . . , -.--
Corollary 2 (Waldman) : Suppose that Eu > E~ and 0U2< 0: . Then for r close enough
to 1, v= > Vp.
Figure 6 shows why Johnson and Waldman reached opposite conclusions about the
predicted occupational choice time-path in the Bandit model. Both of were correct of course,
They just used different parameter values,
5. Evidence: Stepping Stone or Bandit?
The most commonly examined evidence is that on the relation between seniority and
wages; and many authors have asked whether it reflects the growth of skills (as occurs in the
Stepping Stone model) or the effect of selection by characteristic of individuals and jobs (as
occurs in the Bandit model) -- e.g., Altonj i and Shakotko (1987), Topel (1981). We will not add
to this, but rather discuss other implications that emerge from our analysis.
5.1 Evidence favorinu the Steu~in~ Stone model:
Safe tasks first. risky ones last: (The Bandit model implies the opposite) A natural
measure of complexity is the steepness of the learning curve. In our model, a novice’s expected
productivity in occupations A and B is 1- 0X2 and 1 - OY2 respectively, whereas his terminal
productivity is equal to unity in both. So the learning curve is steeper in B. At first blush this
may seem counterfactual, since we know that people tend to focus their learning early on in life.
But this is exactly what the Stepping Stone model says -- early jobs prepare us for later j ohs. By
the time he directs his first operation, a Brain Surgeon has held other jobs -- Intern, Surgical
Assistant -- that have prepared him for the task at hand, so that we would not expect much
further learning. [In keeping with this, we found only a modest learning curve in angioplasty
operations in Jovanovic and Nyarko (1995a)]. The same is true of a Chief Executive Officer; by
the time he assumes the position, he has typically held a whole host of jobs. The issue is one of
sample selection -- if we forced a novice out of medical schools to perfo~ actual operations on
the brain, we would see a very steep learning curve!
A different explanation for why people do safe jobs first is based on incomplete insurance
markets, and risk-aversion: People get richer as they age, and they can tolerate the greater risk.
Moreover, in evaluating any hypothesis on the effects of age, one must include time effects that
control for technological change and other forces that are unrelated to the personal growth of
22
human capital that weare discussing here, Oneuseful dataset isthecross-sectional survey of
Indian farmers studied by Foster andRosensweig(1995),
Share of Rabi-Season Acreage Planted to Different Crops by AgeofHead
Age Wheat Pulses Other*
25 0.25 0.42 0.21
35 0.42 0.18 0.27
45 0.45 0.18 0.25
55 0.42 0.20 0.28
65 0.57 0.14 0.20
75 0.60 0.16 0.16
* Includes sorghum, corn, other cereals, oilseeds, sugarcane, and cotton. A residual category
consisting primarily of vegetables is not shown,
share of wheat = 0.25 + 0.004 Age + 0.0004 Land Adj. R2 = 0.02, n=4,118(4.21) (4.21) (3.34)
Table 2
We can distinguish the risk-aversion/imperfect capital markets hypothesis from ours with the
Foster and Rosensweig data. The gave us the information in Table 2 and the accompanying
regression. The data are from the 1970-71 round of the National Council of Applied Economic
Research Additional Rural Incomes Survey. They describe rural households in India from a national
probability survey begun in 1968-69. From the viewpoint of Table 2 in which there is a 50-year
variation in age, these are essentially cross-section data. For 4826 households, the data provide
information by crop on acreage planted during the two main agricultural seasons in India (Kharif and
Rabi) as well as detailed information on the household including age of the household head and
landholdings which are correlated with wealth. These can be used to proxy for any wealth effect
on choice of crop. The table and regression focuses on the Rabi season crops. Table 2 shows that
young farmers plant more pulses (a type of lentil) and later on in life shift toward planting wheat.
23
The regression shows that this is not caused by a wealth effect. A form of expertise that is
transfened from pulses to wheat is knowing how to deal with external markets: Wheat is more likely
to be a cash crop and hence riskier than pulses, and pulses may offer a more gradual introduction
to external markets.
Is it reasonable to associate costliness of mistakes with the concept of complexity of an
activity? The learning curve is steeper in B in the sense that the ratio of ideal to initial expected
performance, I/oY 2, is higher. We find it natural to think of B as the more complex activity.
Complexity goes hand in hand with steepness of learning if one followsKremer(1993) and defines
a complex activity as embodying a greater number of tasks. [See equation (12) of Jovanovic and
Nyarko (1995a) for a learning curve formula when there are N “stages to an activity. This formula
exhibits greater steepness when N is larger]. If we accept this, then our model implies that simple
activities should precede complex ones. In support of this implication, Wilk and Sackett (1995) find
that people do hold more complex jobs as they age. Their measure is Gottfredson’s (1986)
Occupational Aptitude Patterns Map, an index of cognitive ability needed to perfom the activities
on a job: Vocabulary, reading, math, perceptual speed, and so on. The index assigns each job a
“complexity integer” on a scale of 1 to 10. Chemists, physicians, and engineers are 10’s, while yarn
sorters, general laborers, and baker helpers are 1‘s. Wilk and Sackett’s first table shows that in the
NLSY sample, over a span of 12 years, this index of the cohort rose by more than a full category,
from 3,35 to 4.84.
Climbing the occupational ladder: Hall and Kasten (1976), Miller (1984), Sicherman and
Galor (1990), Baker, Gibbs, and Holmstrom (1994), and Biddle and Roberts (1994) all provide
vertical rankings of occupation, based mostly on pay, and they all find that workers gradually move
up the occupational ladder, This also means that their earnings tend to rise, as they also do in the
Stepping Stone model. Earnings rise with experience in the Bandit model too, but there workers
would, according to Corollary 1, want to start at the top of the ladder right away, because the highest
paid occupations also tend to be more risky.
The transfer of skills from blue-collar work to white collar work: In the NLSY data Neal
(1995) finds one intra-firm occupation change for every six spells of full-time employment. For
white collar workers, a large fraction of these moves are from a given occupation into management.
For blue collar workers, most changes are movements between crafismen, operative, and laborer,
but still with significant movement into management. The Stepping Stone model says that this is
because skills are transferred from blue collar work to white collar work. This claim gets support
from Keane andWolpin(1994) who study the impact on wages of prior experience of people that
24
have moved in both directions. They find (Table 5) that blue collar experience “raises” white collar
wages by more than white collar experience raises blue collar wages,
5.B Evidence favoring the Bandit model:
Complexity and innate ability: Wilk and Sackett (1995) and Wilk, Desmaris, and Sackett
(1995) also found a definite tendency for individuals with higher cognitive ability (an innate trait)
to gravitate towards the more complex activities over time, suggesting that a filtering or screening
process goes on with the accumulation of experience. In keeping with this, Murnane, Levy, and
Willett (1995) find that cognitive skills become increasingly related to wages as people age. These
findings strongly favor the bandit model, since in the Stepping Stone model people are all ex ante
the same.
Sociologists’ studies of occupational ladders: Sociologists have studied occupational ladders
from the point of view of a career path, but their evidence does not seem to show a rise in the
complexity of work as people age. For example, Wilensky (1961) defines a career as a succession
of related jobs, arranged in a hierarchy of prestige through which persons move in an ordered, more
or less predictable sequence. He divides job transitions as orderly if the skills and experience
gained on one job are directly functional to performance in the next job and jobs are arranged in a
hierarchy of prestige. He then defines a disorderly career as one in which less than a fifth of the
work life is in fictionally related, hierarchically ordered jobs. Wilensky’s Table 1 (p, 526) shows
that 63 YO of the of the sample has orderly careers. So clearly there are occupational ladders, and this
perhaps favors the Stepping Stone model. But the study of complexity of work is not supportive of
the that model. The complexity scale most widely used by Sociologists is in the Dictionary of
Occupational Titles (DOT), put out by the Department of Labor scores occupations based on
measures of complexity. There are three dimensions, dealing with the complexity of relationships
to data, people, or things. Cohn and Schooler (1983) have linked the concept of a career to the
concept of complexity; the latter being based on the DOT and on their own surveys of the people
working in an occupation. Among the 3100 men in their sample, ~. 111, Figure 5.2]. shows that
the correlation between complexity in 1964 and complexity in 1974 is 0.77. However, they do not
show that complexity of work increases over one’s lifetime. Finally, using data from a survey of the
Institute for Social Research done in 1967, Baron, and Bielby (1982) also get mixed results on the
relation between age and complexity.
25
6. The Hybrid Model
A person’s productivity depends on innate traits and on the amount of acquired expertise, and
a future study that tries to quantify how important each force is will need a hybrid of the Stepping
Stone and Bandit models. We now sketch one such hybrid. The production functions now are:
qA = a[l - (X - ZA)2] and % = P[l - (y - z~)’] . (7)
The variables are defined as before. Assume ~ signal confusion, so that a period’s experience in
growing apples reveals both a ~ x fully, and similarly, a period’s experience with banana
growing reveals ~ and y. Further, the vector (x,y) is independent of the vector (u,~).
6.1. Symmetric occupations with suecific learning.
The simplest case is one of complete symmetry between the two occupations. Assume that
u and ~ are independent and normally and identically distributed, with mean k and with variance
s 2. Moreover, assume that x and y are also independent (so that p = O and learning is
occupation-specific) and identically and normally distributed, with variance equal to 02. In this
case, ~ the movement is due to mismatch, and learning acts as a deterrent, since it creates
occupation-specific knowledge. Suppose, moreover, that p = 1. Both occupations look identical
ex-ante. Someone who has entered A when young will switch to B if and only if
a<k(l -02). (8)
The left-hand side of this inequality is the person’s earnings in A should he decide to stay there,
and the right-hand side gives his expected earnings should he switch to B. Let @ denote the
standard cumulative normal integral. Since a similar inequality applies to those who enter B when
young, the economy wide rate of movement is equal to the probability that the above inequality is
satisfied, and this equals 0( - ko 2/s). Here, a is an index of occupational-specific human capital
and is decreasing in o. Further,
lim 0( - ku 2/s) = %, (9)0“0
a result that we would have expected given symmetry and the absence of switching costs.
6.2. S mmetric and~ = 1. and UX2<UY2.
In this case, we expect some Stepping Stone movement, or at least we would get it if u and
~ were degenerate. Let p be such that the mean productivity of novices is the same in the two
26
occupations. That is, p satisfies
k(l - 0X2) = pk(l - u;). (lo)
Since p = 1, oneperiod's experience teaches theworker howtopetiom bothjobs perfectly. But
a and ~ are still independent. For the sake of illustration (this is ~ an equilibrium argument),
assume that young workers are, as in 6.1., equally divided between occupations A and B. Then
occupational switchers consist of those who when young worked in A, but who found out that their
a < pk, and those who when young worked in B, but found that ~ < p-’k. The rate of
movement now is
10[
(0; - 0X2)k
— 12 (1 - 0;) s
+1
@[-(o; - o:) k
— 12 (1 - u:) s
(11)
In the absense of Stepping Stone forces (i.e., when 0X2= ~2 =0), since IXand ~ are symmetric
movement due to Bandit forces cancel out. In particular, the two terms in(11 ) are equal to each
other and are both equal to 1/4. Now introduce Stepping Stone forces (i.e., UX2< 0Y2). Then the
first term exceeds the second. The difference between the two terms can loosely be called Stepping
Stone movement.
7. Discussion and Extensions
7.1. Nonlinear ~ricin~ of efficiency. and absolute advantage.
In the Stepping Stone model we assumed that in A and in B there is a constant equilibrium
price (1 and p respectively) per efficiency unit. If I produce twice as much q*, my equilibrium
reward will be twice as large. Since everyone in the Stepping Stone model is ex ante the same, no
one has an absolute advantage over anyone else. In the Bandit model, we normalized both prices
to (1, 1). In the Bandit model some workers also have an absolute advantage. Nevertheless,
because prices are linear in skill in both models, to determine the equilibrium assignments, only the
h of expected productivities in the second period, and only the ti of expected lifetime incomes
of career paths matters in the first period. In this sense assignments are based entirely on
comparative advantage. But this assumption is inappropriate in situations in which one can not
substitute the quantity of service for the auality of service. For example, suppose that for some
27
reason a restaurant must order its apples from a single source so that every restaurant can deal with
only one apple producer. There is no reason why an apple producer that is twice as efficient should
get twice the reward. Sattinger (1975), Rosen(1981 ), and Becker(1981) discuss settings in which
the linearity fails. In their settings skill is exogenous. In ours skill is acquired though
experience, (Kremer (1993, section 2) discusses skill-augmenting education in such a model). Still,
our analysis should apply to this setting as well. Instead of being qA and q~, the rewards to the
two activities would equal pA(qA)~ and p~(q~)q~ respectively,
where p*(.) and p~(.) would be two non-linear price finctions for the level of each type of skill.
(In the models of Sattinger, Becker and Kremer, the pi(.) functions are increasing because a more
productive worker is in equilibrium combined with more productive complementary resources). The
spirit of the analysis remains the same.
7.2. ~.
Workers in the model are apriori identical, but it isn’t hard to indicate how one would allow
for ex-ante differences in workers’ characteristics. Even if producers of apples and bananas should
differ in some way, it is natural to maintain the assumption that they all produce the same product,
and that as a result, they all face the same prices. In a model in which information enhances
productivity oust like in the Stepping Stone model)Rosensweig(1995, p. 154) asks how schooling
may affect productivity; he argues that schooling can tighten priors around optimal decisions, and
it can raise the informational content of each use of the technology. And in a Bandit model, Johnson
(1978) and MacDonald (1980) argues that schooling raises earnings by informing workers about
where his optimal assignment is -- that is, schooling tightens priors about comparative advantage.
7.3. Terms of trade and wage differentials.
In the last 20 years U.S. manufacturing employment has shrunk, and the skilled-unskilled
wage differential has increased. Some have examined the possibility that this was caused by growth
of manufacturing productivity abroad (Berman, Bound, and Griliches 1994). A well-documented
transition on a job ladder is the flow of labor from blue collar to white collar work (see section 5.1).
Since A + B is the direction in which labor flows in our model, we can think of A as blue collar
(e.g., manufacturing), and B as white collar (e.g., some services) If the rest of the world learns how
to manufacture more efficiently, the demand for A should fall sop should rise. We should see a
rightward move along the second step on the supply curve in figure 3. We can therefore analyze
the effect of such changes in the world’s productivity through their effect on p.
28
7.4. Internal labor markets.
We have not discussed firms at all, and yet we know that a lot of occupational movement
happens inside firms, and that firms tend to promote from within. From the results of Neal (1995)
we know that there is at least one intra-firm occupation change for every six spells of full-time
employment, and not every such spell represents an occupational change either. So internal
occupational movement is important, and the present model could generate a distinction between
internal and external occupational movement by introducing firms, and some reason why there
would be firm-specific components to the (x, y) vector.
7.5. Valuin~ ste~uin~ stone mobilitv.
In the model of section 2, to what extent does the option to change jobs raise GNP? In the
context of a Bandit model Jovanovic andMoffltt(1990) found the figure to be 8°/0 of GNP. If we
value the two kinds of output at their equilibrium prices -- 1 for apples and p for bananas -- then
when 5 = 1 (so that the gain in lifetime value equals the gain in lifetime output), the number is easy
to calculate. The increased output is just the number of young that move (see Figure 4), multiplied
by the lifetime utility gain associated with moving: VA~ - max{V**,V~~ }.
7.6. The advantages of countrywide occupational diversitv.
The Stepping Stone model and the Bandit model ~ imply that with a richer menu of
occupations to choose from, the worker’s lifetime productivity will be higher. A country that
specializes in either apples or bananas will not be able to afford its nationals the gains from either
kind of mobility -- a disadvantage of specialization.
8. Conclusion
The Stepping Stone model explains why the complexity of work increases over one’s
lifetime. We have used this model to discuss the concept of occupational mobility. In contrast to
the Bandit model, the quality of the technologies is known in the Stepping Stone model; what is
uncertain best to operate the technology. The Stepping Stone model says that activities that are
inforrnationally “close” will form a ladder, but the safer ones should come fi : They are a natural
training ground because in such activities, mistakes imply smaller foregone output. The model also
says that labor will flow from occupations with flatter learning curves and into occupations where
learning curves are steep, as long as learning is sufficiently transferable between occupations.
29
The general message of this paper is however, much broader than the question of
occupational mobility that we have focused on, and it is this: To understand human capital, it helps
to distinguish information about personal traits and about the content of jobs on the one hand, from
information about how to perform a given task on the other. The informational approach to human
capital has so far stressed the first kind of information, information about parameters pertaining to
traits of people and jobs, The Stepping Stone model illustrates that information theory can also
illuminate the process by which people acquire general as well as job-specific skills.
References:
Altonji, J., and Shakotko, R.
(1987) “Do Wages Rise with Job Seniority?” Review of Economic Studies, 54:437-459.
Baker, G., Gibbs, M., and Holmstrom, B.
(1994) “The Internal Economics of the Firm -- Evidence fi-om Personnel Data,” Quarterly
Journal of Economics 109, no. 4:881-920.
Baron, J., and Bielby, W.
(1982) “Workers and Machines: Dimensions and Determinants of Technical
Relations in the Workplace,” American Sociological Review 47 (April): 175-188.
Becker, G.
(198 1) Treatise on the Family. Cambridge: Harvard University Press.
Berman, E., Bound, J. and Griliches, Z.
(1994) “Changes in the Demand for Skilled Labor within U.S. Manufacturing: Evidence
from the Annual Survey of Manufactures,”O uarterlv Journal of Economics 109, no. 2:367-
398.
Biddle, J. and Roberts, K.
(1994) “Private Sector Scientists and Engineers and the Transition to Management,” Journal
of Human Resources 29, no. 1:82-107.
Chow, G.
(1983) Econometrics New York: McGraw-Hill.
Cohn, M. and Schooler, C.
(1983) Work and Personalit Y: An Inauiry into the Im~act of Social Stratification
New Jersey Ablex Publishing Corporation.
Davis, S. and Haltiwanger, J.
(1992) “Gross Job Creation, Gross Job Destruction, and Employment Reallocation,”
Quarterly Journal of Economics 107, no. 3:819-864.
Foster, A. and Rosensweig, M.
30
(1995) Learning by Doing and Learning from Others: Human Capital and Technical Change
in Agriculture,” Journal of Political Economv 6, no. 103: 1176-1209.
Gottfredson, L.
(1986) “Occupational Aptitude Patterns Map: Development and Implications for a Theory
of Job Aptitude Requirements,” Journal of Vocational Behavior 29:254-291.
Hall, R. and Kasten, R.
(1976)’’Occupational Mobility and the Distribution of Occupational Success Among Young
Men,” American Economic Review. Papers and Proceedings 66, no. 2:309-315.
Johnson, W.
(1978) “A Theory of Job Shopping,” Quarterlv Journal of Economics: 261-277
Jovanovic, B. and Moffitt, R,
(1990) “An Estimate of a Sectoral Model of Labor Mobility,” Journal of Political Economy
98, no. 4:827-52.
Jovanovic, B. and Nyarko, Y.
(1995) “A Bayesian Learning Model Fitted to a Variety of Empirical Learning Curves,”
Brookin~s Papers on Economic Activity Macroeconomics Issue, no. 1.
Keane, M. and Wolpin, K.
(1994) “The Career Decisions of Young Men,” unpublished, New York University.
Kremer, M.
(1993) “The O-Ring Theory of Economic Development,” Quarterly Journal of
Economics 108, no. 3:551-576.
Krusell, P., Ohanian, L., Rios-Rull, J. and Violante, G.
(1995) “Capital-Skill Complementarity and Inequality,” unpublished paper, University of
Rochester.
MacDonald, G.
(1980) “Person-Specific Information in the Labor Market,” Journal of Political Economy 88,
no. 3:578-97.
Miller, R.
(1984) “Job Matching and Occupational Choice” Journal of Political Economy 92, no. 6:
1086-1120.
Mincer, J.
(1974) Schooling. Experience and Eamin~s Brookfield, VT: Ashgate Publishing.
Mumane, R., Levy, F. and Willett, J.
(1995) “The Growing Importance of Cognitive Skills in Wage Determination,” NBER
working paper #5076.
Neal, Derek,
(1995) “The Complexity of Job Mobility Among Young Men,” unpublished, University of
Chicago.
O’Flaherty, B. and Siow, A.
(1992) “On the Job Screening, Up or Out Rules, and Firm Growth,” Canadian Journal of
Economics 25, no. 2:346-67.
O~eill, D.
(1995) “Wage Dispersion over the Life Cycle: Implications for Models of Individual Wage
Determination,” unpublished, University of Newcastle upon Tyne.
Prescott, E. C., and Visscher, M.
(1980) “Organization Capital” Journal of Political Economy 88 (June): 446-61.
Rosen, S.
(1972) “Learning and Experience in the Labor Market,” Journal of Human Resources 7:325
-42.
Rosen, S.
(198 1) “Substitution on Production,” Economics.
Rosensweig, M.
(1995) “Why Are There Returns to Schooling?” American Economic Review, Papers and
Proceedings 85, no. 2:153-58.
Roy, A.D.
(195 1) “Some Thoughts on the Distribution of Earnings,” Oxford Economic Papers mew
-3:135-146.Sattinger, M.
(1975) “Comparative Advantage and the Distribution of Earnings and Abilities,”
Econometrics 18:455-68.
Sicheman, N. and Galor, O.
(1990) “A Theory of Career Mobility,” Journal of Political Economy 98, no. 1:145-76.
Topel, R.
(1991) “Specific Capital, Mobility, and Wages: Wages Rise with Job Seniority,” Journal of
Political Economy 99, no. 1: 145-176.
Waldman, M.
(1984) “Job Assignments, Signaling, and Efficiency,” Rand Journal of Economics, 15, no.2
(Summer): 255-270.
Weiss, Y.
(1971) “Learning by Doing and Occupational Specialization,” Journal of Economic Theo~
3:189-98.
32
Wilensky, H.
(1961) “Orderly Careers and Social Participation: the Impact of Work History
on Social Integration in the Middle Class,” American Sociological Review 26:521-39.
Wilk, S,, Desmaris, L. and Sackett P.
(1995), “Gravitation to Jobs Commensurate with Ability: Longitudinal and Cross-Sectional
Tests”, Journal of A~plied Psycholo~v 80, no. 1:79-85.
Wilk, S. and Sackett, P.
(1995) “A Longitudinal Analysis of Ability-Job Complexity Fit and Job Change,”
unpublished, Department of Management, The Wharton School.
33
Appendix
Proof of Proposition 4: Define AEE~ - Ea and let aO E u - Eu , and PO ❑ ~ - E~. From
Lemma 2,
V== Ea + 6EMax{a, E[~ Ia]} = (1+6)Eu + bE[Max {uO,A + duO} ].
Similarly,
VD= (l+6)E~ + 5E[Max {~0, -A + c~O}].
Consider as fixed 0P2and UU2with 0P2> U=2. Consider the expressions above functions of A and
r. Define D(A,r) ❑ V= - VP. Then
D(A,r) = (1+6)A + 6[E[Max {~,, -A + c~O}] - E[Max {aO,A + daO}]]. (12)
Using the symmetry of the normal distribution and the fact that EaO= O it is easy to check
that E[Max {uO,daO}] = Id-1 IEII aO1]/2 = Id-1 Iou/(2n)]’2 = lr]’20P - Ua l/(2n)1’2 , Similarly,
E[Max {~O,c~O}]= Irl’20m- OPl/(2n)1’2 = ~ - ?’2~, (where the last equality follows from r<l and
OP>au), Hence
D(O,r) = {l/(2n)”2} {UP - rl/20~ - Irlf20b- u=I}
({1/(2n)%} {(oP + u.)(1 - r“2)} when r z a~/u~
(13)
It is easy to see that OD2> U.2 implies that
D(O,r) >0 for all r in [0,1) and D(O,l)=O. (14)
Assume for now that all the terms in the definition of D(A,r) above are differentiable with
respect to A. Then it should be clear that \dE[Max {PO,-A + c~O}/~A I s 1 and I~E[Max {~, A
+ duO}]/~AI s 1 so ~D(A,r)/dA z (1+5) -25 = 1-b >0. In particular, D(A,r) is strictly increasing in
A. By replacing derivatives with derivatives from the left and right, it is easy to check that
differentiability is not required for this latter result.
Hence, using the obvious continuity of D in A, there exists for each r in [0,1], a unique
critical value A’(r) of A such that D(A,r)>O for A> N(r) ; D(A’(r),r)=O ; and D(A,r)<O for A<
K(r). From (14), K(r) <0 for all r in [0,1). Finally, from (13), D(O,l) = O. Since M(l) is by
definition the unique value of A such that D(A,l) = O,we conclude that &(l)=O. ■
34
Extending the argument to a general equilibrium context.
There is a representative consumer with income I’ to spend on CAunits of apples, c~ units
of bananas and M units of a composite good M with prices pA,p~ and 1 respectively. The consumer
therefore solves the problem
Max U(CA,CB,M) St. plcl + pzcz + M K I. (15){CA,CB,M}
We suppose the agent has Cobb-Douglas utility
U(CA,CB,M) = CAPCBVM’-P-V .
This results in the demands for apples and bananas given by
CA = ~l/pA ad CB = VI/pB
(16)
(17)
The relative price of bananas with respect to apples is therefore
P - PB/PA=(V/W).(CB/CA )-1 . (18)
which is the relative demand curve of bananas (relative to apples) as a function of the relative price
of bananas. By adjusting the values of p~ and pAwe can arbitrarily adjust the levels of apple and
banana demands keeping the ratio fixed.
In figures 8 and 9 below we draw this demand curve. We draw on the same diagram the
supply curve of bananas relative to apples (and where we ignore the sections where apple production
is zero so relative banana production is infinite). In figure 8 we see that when p=O there is a unique
equilibrium with price p= ~1and relative output of bananas equal to Q~*/QA*= v/(~~l), [where from
section 2, ~1 ~ (1-uX2+5)/(1-uY2+5)].
When p=l, the equilibrium price depends upon the position of the demand curve, and in
particular on the value of v/p. Figure 9 depicts three such demand curves, D, D‘ and D“,
corresponding to relatively low, medium and high values of v/p. These yield equilibrium relative
prices of p=l, p in (1 ,~) and p=~ respectively [where from section 2, ~ = (1-o~)/(1-a~)].
35
P
o QB*IQA*
Fiz. 8: The Relative SuD~lv and Demand Curves of Bananas when D=O
36
p=(l-lJ:)/(1-oy2)
1
supply
I11I
I
I- QB/QA
o QB*IQ; l/(1-~~) Q~**/QA”*
Fig. 9: The Su~~lv Curve and Three Different Demand Curves for Bananas when 0=1.
37
The Farmer’s Intertemporal income/savings decision and Risk Neutrality
In the text we assume that the farmer will maximice lifetime exuected income. We now
show that this assumption is not inconsistent with the preferences described above in this appendix.
Suppose that all consumption of an individual takes place & the output when old has been
received. In particular there is no consumption when young. Suppose that before consumption the
old have income of I (made up of income received when old plus income when young, the latter
perhaps with interest). Recall that in the appendix we have the representative consumer with utility
function CAPC~vM1-P-vover the three goods CA,C~and M with prices pA,p~ and 1. It is easy to check
that the indirect utility function is linear in I. Hence the farmer will be risk neutral in income I.
This means that in making his labor market decisions, the farmer will want to maximize expected
income.
Random versus Non-Random Prices
In deriving the optimal career path of individuals we assume that prices are constant and the
same in each period. So long as we assume a steady state in career choices of young and old, we
shall obtain equal distributions of prices in each period. Next suppose that there is a continuum of
agents so that no individual can influence the price by his own career choice. Then we may replace
the price p with its expected value and arrive at the same conclusions. If in addition the x’s and y’s
of agents are all independent of each other, we can invoke the strong law of large numbers to
conclude that the prices will actually be constant,
Some More Remarks On Lemma 2:
Define
M E {(au2,uP2,COV(a,~) ) ~ R++XR++xR
Then M is the set of parameters that which could
0=2<oP2,and Cov 2(a,~) < au2aP2}. (19)
lefine the variance-covariance matrix of the normal
distribution and which obeys our hypothesis that a=2<uP2. The expressions c and d in Lemma 2 are
functions of the parameter vector m=(o~,u~, Cov (u,~) ) ~ M. For emphasis we shall index these
by m so that c(m)= [Cov (a,~)/0P2 ] and d(m): [Cov (u,~)/a=2 ] , respectively.
Lemma 2’:
a. (i) For all m ~ M, Ic(m) I<1; and (ii) V c E R such that Ic I<1 there exists an m ● M
such that c(m)=c.
b. (ii) For all d ~ R, there exists an m ~ M such that d(m)=d.
Proofi (a) (i) For m~M, c(m)2 = [COV2(a,~)/oP4 ] s a=2/uP2<1, so part a.i follows. (ii) Fix any c
38
such that IcI<l anddefinem=(o .2,uP2, Cov (a,~)) as follows: Set 0U2= & ; define %2 to be any
number in (0=2,1) and define Cov (a,~) by COV2(a,~) = C2%4 and sgn Cov (u,~) = sgn c. Then
clearly U=2< aP2 ; and COV2(~>~)/~a2~p2= ~p2<1” Hence m EM. Also, c(m)2 E COV2(u,~)/0P4 = C2
so c(m) = c.
(ii) Fix any deR. Define m=(o.2,aP2, COV(~,~)) as follows: ‘efine ‘~2 = ‘2’ ‘et “2’0 be any
number in (0,min{d2,1 }), and define Cov (u,~) by COV2(u,~) = d2aa4 and sgn Cov (u,~) = sgn d.
Then clearly 0=2< o;; and COV2(a,~)/o~o~ = 0$<1. Hence m EM. Also, d(m)2 ~ COV2(U,~)/0 ~
= d2 so d(m) = d. ■
39